* Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2014, Intel Corporation.
+ * Copyright (c) 2012, 2015, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
struct list_head *next;
struct list_head cull;
- LASSERT(the_lnet.ln_init);
-
/* NB: use lnet_net_lock(0) to serialize operations on test peers */
if (threshold != 0) {
/* Adding a new entry */
}
unsigned int
-lnet_iov_nob (unsigned int niov, struct iovec *iov)
+lnet_iov_nob(unsigned int niov, struct kvec *iov)
{
unsigned int nob = 0;
EXPORT_SYMBOL(lnet_iov_nob);
void
-lnet_copy_iov2iov (unsigned int ndiov, struct iovec *diov, unsigned int doffset,
- unsigned int nsiov, struct iovec *siov, unsigned int soffset,
- unsigned int nob)
+lnet_copy_iov2iov(unsigned int ndiov, struct kvec *diov, unsigned int doffset,
+ unsigned int nsiov, struct kvec *siov, unsigned int soffset,
+ unsigned int nob)
{
/* NB diov, siov are READ-ONLY */
unsigned int this_nob;
EXPORT_SYMBOL(lnet_copy_iov2iov);
int
-lnet_extract_iov (int dst_niov, struct iovec *dst,
- int src_niov, struct iovec *src,
- unsigned int offset, unsigned int len)
+lnet_extract_iov(int dst_niov, struct kvec *dst,
+ int src_niov, struct kvec *src,
+ unsigned int offset, unsigned int len)
{
/* Initialise 'dst' to the subset of 'src' starting at 'offset',
* for exactly 'len' bytes, and return the number of entries.
}
EXPORT_SYMBOL(lnet_extract_iov);
-#ifndef __KERNEL__
-unsigned int
-lnet_kiov_nob (unsigned int niov, lnet_kiov_t *kiov)
-{
- LASSERT (0);
- return (0);
-}
-
-void
-lnet_copy_kiov2kiov (unsigned int ndkiov, lnet_kiov_t *dkiov, unsigned int doffset,
- unsigned int nskiov, lnet_kiov_t *skiov, unsigned int soffset,
- unsigned int nob)
-{
- LASSERT (0);
-}
-
-void
-lnet_copy_kiov2iov (unsigned int niov, struct iovec *iov, unsigned int iovoffset,
- unsigned int nkiov, lnet_kiov_t *kiov, unsigned int kiovoffset,
- unsigned int nob)
-{
- LASSERT (0);
-}
-
-void
-lnet_copy_iov2kiov (unsigned int nkiov, lnet_kiov_t *kiov, unsigned int kiovoffset,
- unsigned int niov, struct iovec *iov, unsigned int iovoffset,
- unsigned int nob)
-{
- LASSERT (0);
-}
-
-int
-lnet_extract_kiov (int dst_niov, lnet_kiov_t *dst,
- int src_niov, lnet_kiov_t *src,
- unsigned int offset, unsigned int len)
-{
- LASSERT (0);
-}
-
-#else /* __KERNEL__ */
unsigned int
-lnet_kiov_nob (unsigned int niov, lnet_kiov_t *kiov)
+lnet_kiov_nob(unsigned int niov, lnet_kiov_t *kiov)
{
unsigned int nob = 0;
EXPORT_SYMBOL(lnet_kiov_nob);
void
-lnet_copy_kiov2kiov (unsigned int ndiov, lnet_kiov_t *diov, unsigned int doffset,
- unsigned int nsiov, lnet_kiov_t *siov, unsigned int soffset,
- unsigned int nob)
+lnet_copy_kiov2kiov(unsigned int ndiov, lnet_kiov_t *diov, unsigned int doffset,
+ unsigned int nsiov, lnet_kiov_t *siov, unsigned int soffset,
+ unsigned int nob)
{
/* NB diov, siov are READ-ONLY */
unsigned int this_nob;
EXPORT_SYMBOL(lnet_copy_kiov2kiov);
void
-lnet_copy_kiov2iov (unsigned int niov, struct iovec *iov, unsigned int iovoffset,
+lnet_copy_kiov2iov (unsigned int niov, struct kvec *iov, unsigned int iovoffset,
unsigned int nkiov, lnet_kiov_t *kiov, unsigned int kiovoffset,
unsigned int nob)
{
EXPORT_SYMBOL(lnet_copy_kiov2iov);
void
-lnet_copy_iov2kiov (unsigned int nkiov, lnet_kiov_t *kiov, unsigned int kiovoffset,
- unsigned int niov, struct iovec *iov, unsigned int iovoffset,
- unsigned int nob)
+lnet_copy_iov2kiov(unsigned int nkiov, lnet_kiov_t *kiov, unsigned int kiovoffset,
+ unsigned int niov, struct kvec *iov, unsigned int iovoffset,
+ unsigned int nob)
{
/* NB kiov, iov are READ-ONLY */
unsigned int this_nob;
EXPORT_SYMBOL(lnet_copy_iov2kiov);
int
-lnet_extract_kiov (int dst_niov, lnet_kiov_t *dst,
- int src_niov, lnet_kiov_t *src,
- unsigned int offset, unsigned int len)
+lnet_extract_kiov(int dst_niov, lnet_kiov_t *dst,
+ int src_niov, lnet_kiov_t *src,
+ unsigned int offset, unsigned int len)
{
/* Initialise 'dst' to the subset of 'src' starting at 'offset',
* for exactly 'len' bytes, and return the number of entries.
}
}
EXPORT_SYMBOL(lnet_extract_kiov);
-#endif
void
lnet_ni_recv(lnet_ni_t *ni, void *private, lnet_msg_t *msg, int delayed,
unsigned int offset, unsigned int mlen, unsigned int rlen)
{
unsigned int niov = 0;
- struct iovec *iov = NULL;
+ struct kvec *iov = NULL;
lnet_kiov_t *kiov = NULL;
int rc;
lnet_finalize(ni, msg, rc);
}
-void
+static void
lnet_setpayloadbuffer(lnet_msg_t *msg)
{
lnet_libmd_t *md = msg->msg_md;
void
lnet_prep_send(lnet_msg_t *msg, int type, lnet_process_id_t target,
- unsigned int offset, unsigned int len)
+ unsigned int offset, unsigned int len)
{
msg->msg_type = type;
msg->msg_target = target;
msg->msg_hdr.payload_length = cpu_to_le32(len);
}
-void
+static void
lnet_ni_send(lnet_ni_t *ni, lnet_msg_t *msg)
{
void *priv = msg->msg_private;
lnet_finalize(ni, msg, rc);
}
-int
+static int
lnet_ni_eager_recv(lnet_ni_t *ni, lnet_msg_t *msg)
{
int rc;
}
/* NB: caller shall hold a ref on 'lp' as I'd drop lnet_net_lock */
-void
+static void
lnet_ni_query_locked(lnet_ni_t *ni, lnet_peer_t *lp)
{
cfs_time_t last_alive = 0;
int alive;
cfs_time_t deadline;
+ LASSERT (lnet_peer_aliveness_enabled(lp));
+
/* Trust lnet_notify() if it has more recent aliveness news, but
* ignore the initial assumed death (see lnet_peers_start_down()).
*/
/* NB: returns 1 when alive, 0 when dead, negative when error;
* may drop the lnet_net_lock */
-int
+static int
lnet_peer_alive_locked (lnet_peer_t *lp)
{
- cfs_time_t now = cfs_time_current();
- bool query;
+ cfs_time_t now = cfs_time_current();
- if (!lnet_peer_aliveness_enabled(lp))
- return -ENODEV;
+ if (!lnet_peer_aliveness_enabled(lp))
+ return -ENODEV;
- if (lp->lp_last_query == 0) {
- query = true;
- } else {
- /* Peer appears dead, but we should avoid frequent NI queries
- * (at most once per ni_query_interval seconds). */
- static const int ni_query_interval = 1;
- cfs_time_t next_query;
-
- next_query = cfs_time_add(lp->lp_last_query,
- cfs_time_seconds(ni_query_interval));
- query = cfs_time_aftereq(now, next_query);
+ if (lnet_peer_is_alive(lp, now))
+ return 1;
+
+ /* Peer appears dead, but we should avoid frequent NI queries (at
+ * most once per lnet_queryinterval seconds). */
+ if (lp->lp_last_query != 0) {
+ static const int lnet_queryinterval = 1;
+
+ cfs_time_t next_query =
+ cfs_time_add(lp->lp_last_query,
+ cfs_time_seconds(lnet_queryinterval));
+
+ if (cfs_time_before(now, next_query)) {
+ if (lp->lp_alive)
+ CWARN("Unexpected aliveness of peer %s: "
+ "%d < %d (%d/%d)\n",
+ libcfs_nid2str(lp->lp_nid),
+ (int)now, (int)next_query,
+ lnet_queryinterval,
+ lp->lp_ni->ni_peertimeout);
+ return 0;
+ }
}
/* query NI for latest aliveness news */
- if (query)
- lnet_ni_query_locked(lp->lp_ni, lp);
+ lnet_ni_query_locked(lp->lp_ni, lp);
if (lnet_peer_is_alive(lp, now))
return 1;
/* NB 'lp' is always the next hop */
if ((msg->msg_target.pid & LNET_PID_USERFLAG) == 0 &&
- lnet_peer_alive_locked(lp) == 0 &&
- !lnet_msg_is_rc_ping(msg)) { /* send RC ping even for dead router */
+ lnet_peer_alive_locked(lp) == 0) {
the_lnet.ln_counters[cpt]->drop_count++;
the_lnet.ln_counters[cpt]->drop_length += msg->msg_len;
lnet_net_unlock(cpt);
return LNET_CREDIT_OK;
}
-#ifdef __KERNEL__
-lnet_rtrbufpool_t *
+static lnet_rtrbufpool_t *
lnet_msg2bufpool(lnet_msg_t *msg)
{
lnet_rtrbufpool_t *rbp;
return rbp;
}
-int
+static int
lnet_post_routed_recv_locked (lnet_msg_t *msg, int do_recv)
{
/* lnet_parse is going to lnet_net_unlock immediately after this, so it
}
return LNET_CREDIT_OK;
}
-#endif
void
lnet_return_tx_credits_locked(lnet_msg_t *msg)
}
}
-#ifdef __KERNEL__
void
lnet_schedule_blocked_locked(lnet_rtrbufpool_t *rbp)
{
(void)lnet_post_routed_recv_locked(msg, 1);
}
-#endif
void
lnet_drop_routed_msgs_locked(struct list_head *list, int cpt)
lnet_return_rx_credits_locked(lnet_msg_t *msg)
{
lnet_peer_t *rxpeer = msg->msg_rxpeer;
-#ifdef __KERNEL__
lnet_msg_t *msg2;
if (msg->msg_rtrcredit) {
/* It is possible that a user has lowered the desired number of
* buffers in this pool. Make sure we never put back
* more buffers than the stated number. */
- if (rbp->rbp_credits >= rbp->rbp_nbuffers) {
- /* Discard this buffer so we don't have too many. */
+ if (unlikely(rbp->rbp_credits >= rbp->rbp_req_nbuffers)) {
+ /* Discard this buffer so we don't have too
+ * many. */
lnet_destroy_rtrbuf(rb, rbp->rbp_npages);
+ rbp->rbp_nbuffers--;
} else {
list_add(&rb->rb_list, &rbp->rbp_bufs);
rbp->rbp_credits++;
(void) lnet_post_routed_recv_locked(msg2, 1);
}
}
-#else
- LASSERT(!msg->msg_rtrcredit);
- LASSERT(!msg->msg_peerrtrcredit);
-#endif
if (rxpeer != NULL) {
msg->msg_rxpeer = NULL;
lnet_peer_decref_locked(rxpeer);
{
lnet_peer_t *p1 = r1->lr_gateway;
lnet_peer_t *p2 = r2->lr_gateway;
-
- if (p1->lp_ni->ni_peertimeout > 0 &&
- p2->lp_ni->ni_peertimeout > 0) {
- /* if a router has queued bytes but no aliveness update for
- * the last 10 seconds, it could be potentially dead or
- * congested, so we prefer not to choose it even its status
- * is still alive.
- */
- int router_slow = cfs_time_seconds(10);
- bool r1_slow;
- bool r2_slow;
- cfs_time_t now = cfs_time_current();
-
- r1_slow = p1->lp_txqnob != 0 &&
- cfs_time_aftereq(now, p1->lp_last_alive + router_slow);
- r2_slow = p2->lp_txqnob != 0 &&
- cfs_time_aftereq(now, p2->lp_last_alive + router_slow);
-
- if (!r1_slow && r2_slow)
- return 1;
-
- if (r1_slow && !r2_slow)
- return -1;
- }
+ int r1_hops = (r1->lr_hops == LNET_UNDEFINED_HOPS) ? 1 : r1->lr_hops;
+ int r2_hops = (r2->lr_hops == LNET_UNDEFINED_HOPS) ? 1 : r2->lr_hops;
if (r1->lr_priority < r2->lr_priority)
return 1;
if (r1->lr_priority > r2->lr_priority)
return -1;
- if (r1->lr_hops < r2->lr_hops)
+ if (r1_hops < r2_hops)
return 1;
- if (r1->lr_hops > r2->lr_hops)
+ if (r1_hops > r2_hops)
return -1;
if (p1->lp_txqnob < p2->lp_txqnob)
}
LASSERT (lp->lp_ni == src_ni);
} else {
-#ifndef __KERNEL__
- lnet_net_unlock(cpt);
-
- /* NB
- * - once application finishes computation, check here to update
- * router states before it waits for pending IO in LNetEQPoll
- * - recursion breaker: router checker sends no message
- * to remote networks */
- if (the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING)
- lnet_router_checker();
-
- lnet_net_lock(cpt);
-#endif
/* sending to a remote network */
lp = lnet_find_route_locked(src_ni, dst_nid, rtr_nid);
if (lp == NULL) {
lnet_hdr_t *hdr = &msg->msg_hdr;
struct lnet_match_info info;
int rc;
+ bool ready_delay;
/* Convert put fields to host byte order */
hdr->msg.put.match_bits = le64_to_cpu(hdr->msg.put.match_bits);
info.mi_mbits = hdr->msg.put.match_bits;
msg->msg_rx_ready_delay = ni->ni_lnd->lnd_eager_recv == NULL;
+ ready_delay = msg->msg_rx_ready_delay;
again:
rc = lnet_ptl_match_md(&info, msg);
return 0;
case LNET_MATCHMD_NONE:
- if (msg->msg_rx_delayed) /* attached on delayed list */
+ if (ready_delay)
+ /* no eager_recv or has already called it, should
+ * have been attached on delayed list */
return 0;
rc = lnet_ni_eager_recv(ni, msg);
- if (rc == 0)
+ if (rc == 0) {
+ ready_delay = true;
goto again;
+ }
/* fall through */
case LNET_MATCHMD_DROP:
}
CDEBUG(D_NET, "%s: Reply from %s of length %d/%d into md "LPX64"\n",
- libcfs_nid2str(ni->ni_nid), libcfs_id2str(src),
+ libcfs_nid2str(ni->ni_nid), libcfs_id2str(src),
mlength, rlength, hdr->msg.reply.dst_wmd.wh_object_cookie);
lnet_msg_attach_md(msg, md, 0, mlength);
}
CDEBUG(D_NET, "%s: ACK from %s into md "LPX64"\n",
- libcfs_nid2str(ni->ni_nid), libcfs_id2str(src),
+ libcfs_nid2str(ni->ni_nid), libcfs_id2str(src),
hdr->msg.ack.dst_wmd.wh_object_cookie);
lnet_msg_attach_md(msg, md, 0, 0);
{
int rc = 0;
-#ifdef __KERNEL__
if (!the_lnet.ln_routing)
return -ECANCELED;
if (rc == 0)
rc = lnet_post_routed_recv_locked(msg, 0);
-#else
- LBUG();
-#endif
return rc;
}
return ("<UNKNOWN>");
}
}
-EXPORT_SYMBOL(lnet_msgtyp2str);
void
lnet_print_hdr(lnet_hdr_t * hdr)
lnet_parse(lnet_ni_t *ni, lnet_hdr_t *hdr, lnet_nid_t from_nid,
void *private, int rdma_req)
{
+ int rc = 0;
+ int cpt;
+ int for_me;
struct lnet_msg *msg;
- lnet_peer_t *rxpeer;
- lnet_pid_t dest_pid;
- lnet_nid_t dest_nid;
- lnet_nid_t src_nid;
- __u32 payload_length;
- __u32 type;
- int for_me;
- int cpt;
- int rc = 0;
-
- LASSERT(!in_interrupt());
+ lnet_pid_t dest_pid;
+ lnet_nid_t dest_nid;
+ lnet_nid_t src_nid;
+ __u32 payload_length;
+ __u32 type;
+
+ LASSERT (!in_interrupt ());
type = le32_to_cpu(hdr->type);
src_nid = le64_to_cpu(hdr->src_nid);
}
lnet_msg_commit(msg, cpt);
- /* LND just notified me for incoming message from rxpeer, so assume
- * it is alive */
- rxpeer = msg->msg_rxpeer;
- rxpeer->lp_last_alive = rxpeer->lp_last_query = cfs_time_current();
- if (!rxpeer->lp_alive)
- lnet_notify_locked(rxpeer, 0, 1, rxpeer->lp_last_alive);
-
- if (lnet_isrouter(msg->msg_rxpeer) &&
- LNET_NIDNET(src_nid) != LNET_NIDNET(from_nid)) {
- lnet_router_ni_update_locked(msg->msg_rxpeer,
- LNET_NIDNET(src_nid));
- }
/* message delay simulation */
if (unlikely(!list_empty(&the_lnet.ln_delay_rules) &&
int cpt;
int rc;
- LASSERT(the_lnet.ln_init);
LASSERT(the_lnet.ln_refcount > 0);
if (!list_empty(&the_lnet.ln_test_peers) && /* normally we don't */
/* NB handles only looked up by creator (no flips) */
if (ack == LNET_ACK_REQ) {
- msg->msg_hdr.msg.put.ack_wmd.wh_interface_cookie =
+ msg->msg_hdr.msg.put.ack_wmd.wh_interface_cookie =
the_lnet.ln_interface_cookie;
- msg->msg_hdr.msg.put.ack_wmd.wh_object_cookie =
+ msg->msg_hdr.msg.put.ack_wmd.wh_object_cookie =
md->md_lh.lh_cookie;
} else {
- msg->msg_hdr.msg.put.ack_wmd.wh_interface_cookie =
+ msg->msg_hdr.msg.put.ack_wmd.wh_interface_cookie =
LNET_WIRE_HANDLE_COOKIE_NONE;
- msg->msg_hdr.msg.put.ack_wmd.wh_object_cookie =
+ msg->msg_hdr.msg.put.ack_wmd.wh_object_cookie =
LNET_WIRE_HANDLE_COOKIE_NONE;
}
if (getmd->md_threshold == 0) {
CERROR ("%s: Dropping REPLY from %s for inactive MD %p\n",
- libcfs_nid2str(ni->ni_nid), libcfs_id2str(peer_id),
+ libcfs_nid2str(ni->ni_nid), libcfs_id2str(peer_id),
getmd);
lnet_res_unlock(cpt);
goto drop;
int cpt;
int rc;
- LASSERT (the_lnet.ln_init);
LASSERT (the_lnet.ln_refcount > 0);
if (!list_empty(&the_lnet.ln_test_peers) && /* normally we don't */
* keep order 0 free for 0@lo and order 1 free for a local NID
* match */
- LASSERT (the_lnet.ln_init);
LASSERT (the_lnet.ln_refcount > 0);
cpt = lnet_net_lock_current();
if (rnet->lrn_net == dstnet) {
lnet_route_t *route;
lnet_route_t *shortest = NULL;
+ __u32 shortest_hops = LNET_UNDEFINED_HOPS;
+ __u32 route_hops;
LASSERT(!list_empty(&rnet->lrn_routes));
list_for_each_entry(route, &rnet->lrn_routes,
lr_list) {
+ route_hops = route->lr_hops;
+ if (route_hops == LNET_UNDEFINED_HOPS)
+ route_hops = 1;
if (shortest == NULL ||
- route->lr_hops < shortest->lr_hops)
+ route_hops < shortest_hops) {
shortest = route;
+ shortest_hops = route_hops;
+ }
}
- LASSERT (shortest != NULL);
- hops = shortest->lr_hops;
- if (srcnidp != NULL)
- *srcnidp = shortest->lr_gateway->lp_ni->ni_nid;
- if (orderp != NULL)
- *orderp = order;
+ LASSERT(shortest != NULL);
+ hops = shortest_hops;
+ if (srcnidp != NULL)
+ *srcnidp = shortest->lr_gateway->lp_ni->ni_nid;
+ if (orderp != NULL)
+ *orderp = order;
lnet_net_unlock(cpt);
return hops + 1;
}
return -EHOSTUNREACH;
}
EXPORT_SYMBOL(LNetDist);
-
-/**
- * Set the number of asynchronous messages expected from a target process.
- *
- * This function is only meaningful for userspace callers. It's a no-op when
- * called from kernel.
- *
- * Asynchronous messages are those that can come from a target when the
- * userspace process is not waiting for IO to complete; e.g., AST callbacks
- * from Lustre servers. Specifying the expected number of such messages
- * allows them to be eagerly received when user process is not running in
- * LNet; otherwise network errors may occur.
- *
- * \param id Process ID of the target process.
- * \param nasync Number of asynchronous messages expected from the target.
- *
- * \return 0 on success, and an error code otherwise.
- */
-int
-LNetSetAsync(lnet_process_id_t id, int nasync)
-{
-#ifdef __KERNEL__
- return 0;
-#else
- lnet_ni_t *ni;
- lnet_remotenet_t *rnet;
- struct list_head *tmp;
- lnet_route_t *route;
- lnet_nid_t *nids;
- int nnids;
- int maxnids = 256;
- int rc = 0;
- int rc2;
- int cpt;
-
- /* Target on a local network? */
- ni = lnet_net2ni(LNET_NIDNET(id.nid));
- if (ni != NULL) {
- if (ni->ni_lnd->lnd_setasync != NULL)
- rc = (ni->ni_lnd->lnd_setasync)(ni, id, nasync);
- lnet_ni_decref(ni);
- return rc;
- }
-
- /* Target on a remote network: apply to routers */
- again:
- LIBCFS_ALLOC(nids, maxnids * sizeof(*nids));
- if (nids == NULL)
- return -ENOMEM;
- nnids = 0;
-
- /* Snapshot all the router NIDs */
- cpt = lnet_net_lock_current();
- rnet = lnet_find_net_locked(LNET_NIDNET(id.nid));
- if (rnet != NULL) {
- list_for_each(tmp, &rnet->lrn_routes) {
- if (nnids == maxnids) {
- lnet_net_unlock(cpt);
- LIBCFS_FREE(nids, maxnids * sizeof(*nids));
- maxnids *= 2;
- goto again;
- }
-
- route = list_entry(tmp, lnet_route_t, lr_list);
- nids[nnids++] = route->lr_gateway->lp_nid;
- }
- }
- lnet_net_unlock(cpt);
-
- /* set async on all the routers */
- while (nnids-- > 0) {
- id.pid = LNET_PID_LUSTRE;
- id.nid = nids[nnids];
-
- ni = lnet_net2ni(LNET_NIDNET(id.nid));
- if (ni == NULL)
- continue;
-
- if (ni->ni_lnd->lnd_setasync != NULL) {
- rc2 = (ni->ni_lnd->lnd_setasync)(ni, id, nasync);
- if (rc2 != 0)
- rc = rc2;
- }
- lnet_ni_decref(ni);
- }
-
- LIBCFS_FREE(nids, maxnids * sizeof(*nids));
- return rc;
-#endif
-}
-EXPORT_SYMBOL(LNetSetAsync);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff